scenario would require a malicious user to be able to control the model which as
stated above is already equivalent to code execution. In any case, the
recommendation is not to serve models using Eager mode since it also has
performance limitations.

## Multi-Tenant environments

It is possible to run multiple TensorFlow models in parallel. For example,

```
.
├── app
│   ├── __init__.py
│   ├── main.py
└── static
    ├── redoc.standalone.js
    ├── swagger-ui-bundle.js
    └── swagger-ui.css
```

### Sirve los archivos estáticos { #serve-the-static-files }

* Importa `StaticFiles`.
* "Monta" una instance de `StaticFiles()` en un path específico.

{* ../../docs_src/custom_docs_ui/tutorial002_py310.py hl[7,11] *}

```
.
├── app
│   ├── __init__.py
│   ├── main.py
└── static
    ├── redoc.standalone.js
    ├── swagger-ui-bundle.js
    └── swagger-ui.css
```

### Prover os arquivos estáticos { #serve-the-static-files }

* Importe `StaticFiles`.
* "Monte" a instância `StaticFiles()` em um caminho específico.

{* ../../docs_src/custom_docs_ui/tutorial002_py310.py hl[7,11] *}

* Restarts
* Replication (the number of processes running)
* Memory
* Previous steps before starting

We'll see how they would affect **deployments**.

In the end, the ultimate objective is to be able to **serve your API clients** in a way that is **secure**, to **avoid disruptions**, and to use the **compute resources** (for example remote servers/virtual machines) as efficiently as possible. 🚀

  for (int i = 0; i < inputs.size(); i++) {
    input_ids[i] = ToId(inputs[i]);
    input_dtypes[i] = inputs[i]->DataType();
  }
  std::vector<TapeTensor> tape_tensors;
  tape_tensors.reserve(outputs.size());
  for (auto t : outputs) {
    tape_tensors.push_back(TapeTensor(t));
  }
  GradientTape::RecordOperation(
      op_name, tape_tensors, input_ids, input_dtypes,

		return msgp.NilSize
	}
	sz := msgp.ArrayHeaderSize
	for _, v := range j.val {
		sz += v.Msgsize()
	}
	return sz
}

// Value returns the underlying value.
// Regular append mechanics should be observed.
// If no value has been set yet, a new array is created.
func (j *Array[T]) Value() []T {
	if j.val == nil {
		j.val = j.p.newA(10)
	}
	return j.val
}

// Append a value to the underlying array.

E esses modelos estão compartilhando muitos dos dados e duplicando nomes e tipos de atributos.

Nós poderíamos fazer melhor.

Podemos declarar um modelo `UserBase` que serve como base para nossos outros modelos. E então podemos fazer subclasses desse modelo que herdam seus atributos (declarações de tipo, validação, etc.).

Toda conversão de dados, validação, documentação, etc. ainda funcionará normalmente.

```
.
├── app
│   ├── __init__.py
│   ├── main.py
└── static
    ├── redoc.standalone.js
    ├── swagger-ui-bundle.js
    └── swagger-ui.css
```

### Statik dosyaları servis edin { #serve-the-static-files }

* `StaticFiles` içe aktarın.
* Belirli bir path'te bir `StaticFiles()` instance'ını "mount" edin.

{* ../../docs_src/custom_docs_ui/tutorial002_py310.py hl[7,11] *}

func (m *Manager) AddToMux(router *mux.Router, authReq func(r *http.Request) error) {
	router.Handle(m.routePath, m.Handler(authReq))
}

// Handler returns a handler that can be used to serve grid requests.
// This should be connected on RoutePath to the main server.
func (m *Manager) Handler(authReq func(r *http.Request) error) http.HandlerFunc {
	return func(w http.ResponseWriter, req *http.Request) {

    public static final int FILE_OPEN_REQUIRING_OPLOCK = 0x10000;
    /**
     * Any open of this file cannot be exclusive
     */
    public static final int FILE_DISALLOW_EXCLUSIVE = 0x20000;
    /**
     * Reserve an opportunistic lock filter on the open
     */
    public static final int FILE_RESERVE_OPFILTER = 0x100000;
    /**
     * Open a reparse point and bypass normal reparse point processing
     */